Wide Receiver Projections: Fun With Air Yards and WR Efficiency
Quality of Opportunity
Opportunity is the lifeblood of fantasy football value. There are two types of opportunity, quantity of opportunity and quality of opportunity. Of the two, typically the efforts of fantasy analysts and players are devoted towards analyzing quantity of opportunity (volume), and rightfully so. Still, we can gain an extra edge by evaluating quality of opportunity in addition to identifying volume to better create wide receiver projections.
We generally see quality of opportunity analysis associated with running backs. It’s easier in that realm where identifying goal-line/red-zone touches and work in the passing game along with game script, offensive line, and team scoring expectations is pretty standard. We’re seeing more advanced work in this areas as well. For example, Josh Hermsmeyer has been doing some excellent work on Twitter this offseason looking at stability rates of YPC (yards per carry) and account for MIB (men in the box). That may lean a bit towards real-life application over fantasy, but you get the idea.
What better explains YPC when you have both total number of defenders in the box, and relative blocking advantage data?
Min number of carries at each combination of MIB and advantage filtered to 25 or more att.
The number of MIB explains ~80% of YPC by itself. pic.twitter.com/FmfixWH8ki
— Josh Hermsmeyer (@friscojosh) June 6, 2018
In the receiving game, this analysis is done but it’s not as commonplace. The advantages of quality of opportunity aren’t as easy to parse out for receivers. Goal line work doesn’t matter *as* much as it does for RBs, and all WRs are pass catchers, so we can’t differentiate them that way.
At DailyRoto, we have customizable wide receiver projections that allow for a large amount of transparency in how our projections work. For receivers, the efficiency metrics we use are yards per target (YPT) and catch rate. We also project market share of receiving TDs, which on the back end we can get to by projecting a player’s TD rate (touchdowns per targets).
As Drew Dinkmeyer and I worked on the first round of projections for our customers, I wanted to approach setting these baselines in a different and better way than we have previously.
So, how was I going to more accurately create YPT, catch rate, and TD rate baselines for pass catchers? Inspired somewhat by Mike Clay’s launch of his OFP statistic (Opportunity-adjusted Fantasy points), I had an idea.
Okay – Today is a huge day. I created a new stat and it's arrived.
First there was aDOT, then OTD.
Today, I give you OFP – Opportunity-adjusted Fantasy points. One stat that shows a players expected fantasy point total based only on opportunity.
— Mike Clay (@MikeClayNFL) August 10, 2018
My idea was simple in theory. I wanted to create a model that looked at the starting yard line of every target and the depth of target (air yards) of each of those targets. The idea being that for each general combination of starting yard line and air yards there would be a standard expectation for yards, catch rate, and TD rate. For example, a target 70 yards away from the end zone with 20 air yards is going to have a much higher yards per target expectation, but much lower TD rate expectation, than a target from the 1-yard line with 1 air yard.
This would allow me to have a quick snapshot of a receiver’s quality of opportunity from the previous season, agnostic of anything else. Based on where these targets are coming from on the field and the average depth of these targets, what would we expect the receiver’s baselines to be?
Thankfully, Ron Yurko selflessly made play by play data available dating back to 2009: https://ryurko.github.io/nflscrapR-data/. Using this data, I trained a k-nearest neighbors model on every target since 2009 to predict YPT, catch rate, and TD rate of a target based only on its starting yard line and air yards.
Here are the out of sample r-squared values for each category:
In other words, our simple model explains 8.7% of the variance in YPT, 9.9% of the variance in Catch Rate, and 20.9% of the variance in TD rate.
Based on this model, here were the leaders in expected YPT, catch rate, and TD rate among WRs/TEs with at least 50 targets last season:
|Player||Targets||Expected YPT||Actual YPT|
|Player||Targets||Expected Catch Rate||Actual Catch Rate|
|Player||Targets||Expected TD Rate||Actual TD Rate|
-Generally speaking, the leaders in each category seem to make sense intuitively.
-Systemically, the average Expected YPT and Expected TD Rates line up with the Actual YPT and TD rates. However, we seem to be under projecting catch rate, which is pretty clear simply comparing the expected catch rates and actual catch rates of the leaders above.
-Keep in mind this is what we’d expect from these players moving forward *if they were to receive the same opportunities*. The model isn’t saying Jimmy Graham is likely to post a 10.8% TD rate again. The model is saying that the 10.8% TD rate that Graham posted last season was in line with what we’d expect based on his actual opportunities.
Of course, players don’t exist in a vacuum where only starting yard line and air yards matter. Quality of QB play, quality of opposing defenses, and receiving skill obviously play a big role as well. As far as the former two are concerned, on a weekly basis at DailyRoto we will project a team’s aggregate expected YPT, catch rate, and TD rate utilizing factors like QB play and quality of opposing defense. We will then scale the WRs, according to their individual baselines, to meet those team set targets and set final wide receiver projections.
We can, and do, use WR skill to set those individual baselines before they are scaled, however. In an effort to make the 2017 expected benchmarks for each play more accurate, I looked at the career data (2009 first year of data set) of each receiver to determine a regressed average over/under expectation for each metric. This was a quick and dirty way of seeing by how much a receiver beats our model’s expectations on average. I call these skill adjustments, although admittedly if doing these correctly you would additionally need to account for historic QB play, strength of opponents, aging curves, and league-wide trends.
I then applied these skill adjustments to get skill-adjusted expectations for each category, simply taking the model’s expectations and then adding each player’s historical over or underperformance in each category.
|Player||Targets||SA Expected YPT||Actual YPT|
|Player||Targets||SA Expected Catch Rate||Actual Catch Rate|
|Player||Targets||SA Expected TD Rate||Actual TD Rate|
You do see some of the flaws in this approach. DeSean Jackson and Antonio Gates are examples of older players who missed their expected marks by a wide margin. Their skill adjustments need to take into account their age. Julius Thomas’ TD rate skill adjustment is driven by time in Denver playing in Peyton Manning, and not necessarily his actual skill.
Tyreek Hill and Doug Baldwin Dominance
So how is this useful? First off, I’d be remiss if we went any further without appreciating how doggone good Doug Baldwin is and the scorching start to his career Tyreek Hill is having. Throwing out players with less than 150 career targets, of the WRs with 50-plus targets last season here are the leaders in career +/- over expectation for all three categories. You will sense a common theme.
|Player||Career YPT Over Expected|
|Player||Career Catch Rate Over Expected|
|Tyreek Hill||13.3% (percentage points)|
|Player||Career TD Rate Over Expected|
|Tyreek Hill||2.3% (percentage points)|
Hill and Baldwin are the only WRs that show up on more than one list, and they show up on all three! I’m not sure which is more impressive – Hill leading all of these categories or Doug Baldwin sustaining this over his 677 targets (192 for Hill).
Secondly, by looking at over/underachievers from the year before we can spot some obvious regression candidates.
It’s possible that JuJu Smith-Schuster is the next elite PIT WR, but more than likely he will have a tough time repeating anywhere close to last year’s efficiency. Even if we give Smith-Schuster some skill credit for what he achieved last year, our model has him as the luckiest receiver in terms of YPT and second luckiest in terms of catch rate.
Another WR that was Top 2 in YPT and catch rate luck was Tedd Ginn. While he will likely regress, it is important to note how dramatic of an effect QB play can have on a WR as Ginn benefitted a ton by playing in the Drew Brees-led Saints offense.
On the flip side, someone like Alshon Jeffery was among the unluckiest receivers in both YPT and catch rate (a little lucky in TD rate). If healthy, we could see some much bigger stat lines from him this season.
Thirdly, this exercise goes to show you just how important *earning* high quality targets is. We haven’t seen Antonio Brown or Julio Jones much, if at all, on any of the above lists. Both of these elite WRs have career +/- numbers in YPT and catch rate that beat out the average numbers of 2017’s qualified WRs. However, it’s not by as much as you’d expect. Their talent resides in consistently earning high quality targets.
Both Brown (seventh) and Jones (14th) were ranked in the Top 20 in skill-adjusted expected YPT last season, receiving 172 and 171 targets respectively. The other 18 members of the Top 20 averaged 107 targets.
It is interesting to note that Brown and Jones had clearly better quality of opportunities than DeAndre Hopkins, so Hopkins is either going to have to continue to out target Brown and Jones or see much better opportunities if he hopes to sniff last year’s production.
If you want some more random observations on this data, you can check out my Twitter thread:
Not adjusted enough for QB skill / age curves, but spent some more time modeling 2017 YPT, Catch Rate, TD Rate expectations for players based on where targets occurred and aDOT.
Here's some observations (min 50 targets): https://t.co/9f7UFgJPCj
— Michael Leone (@2Hats1Mike) August 14, 2018
Making use of this model both with and without skill adjustments and manually accounting for some of its weaknesses that I’ve touched on (age, QB play), we hope our WR efficiency baselines are better than ever entering the season, and we now have a mechanism to more accurately gauge how a receiver’s role in-season may be changing from how they’ve been used the past.
Ultimately, though, I thought this was a cool exercise that gleaned useful information in a variety of ways – identifiying regression candidates, appreciating the efficiency of Tyreek Hill and Doug Baldwin and high quality volume of Antonio Brown and Julio Jones, and learning more about measuring the quality of WR opportunity. Perhaps we aren’t left with a high powered model that has completely earned our trust, but the process was fun and education for me. Hopefully, it was for you as well!
If you feel like you have a great handle on projecting WR opportunity, whether it’s volume or efficiency, one of the cool parts of our DailyRoto lineup optimizer is the ability to customize your own baselines if you disagree with ours.